Abstract: Throughing the study the neural network can give a output which match the expectation output, if use it in traffic flow forecast may achieve the good effect.The traditional BP algorithm has some limitations such as the learning convergence speed is very slow and the learning process is dependent on the learning rate.In view of the limitations, this paper presents a BP neural network based on searching for more adaptive learning rate algorithm.The traffic flow forecasts simulation results show that the new algorithm is less dependent on the learning rate than the traditional BP algorithm.